US20060150127A1 - Method of achieving timing closure in digital integrated circuits by optimizing individual macros - Google Patents
Method of achieving timing closure in digital integrated circuits by optimizing individual macros Download PDFInfo
- Publication number
- US20060150127A1 US20060150127A1 US11/296,780 US29678005A US2006150127A1 US 20060150127 A1 US20060150127 A1 US 20060150127A1 US 29678005 A US29678005 A US 29678005A US 2006150127 A1 US2006150127 A1 US 2006150127A1
- Authority
- US
- United States
- Prior art keywords
- optimization
- timing
- macros
- macro
- objective function
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Abandoned
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/30—Circuit design
- G06F30/32—Circuit design at the digital level
- G06F30/33—Design verification, e.g. functional simulation or model checking
- G06F30/3308—Design verification, e.g. functional simulation or model checking using simulation
- G06F30/3312—Timing analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/30—Circuit design
- G06F30/32—Circuit design at the digital level
- G06F30/327—Logic synthesis; Behaviour synthesis, e.g. mapping logic, HDL to netlist, high-level language to RTL or netlist
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/30—Circuit design
- G06F30/32—Circuit design at the digital level
- G06F30/33—Design verification, e.g. functional simulation or model checking
- G06F30/3315—Design verification, e.g. functional simulation or model checking using static timing analysis [STA]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2119/00—Details relating to the type or aim of the analysis or the optimisation
- G06F2119/12—Timing analysis or timing optimisation
Definitions
- This invention relates to the design, and automation thereof, of high-performance digital integrated circuits.
- the invention is particularly directed to the problem of obtaining timing closure of entire integrated circuits or functional units of an integrated circuit by optimizing or tuning individual macros that contitute the functional unit or integrated circuit.
- IBM® is a registered trademark of International Business Machines Corporation, Armonk, N.Y., U.S.A. Other names may be registered trademarks or product names of International Business Machines Corporation or other companies.
- Achieving timing closure of high-performace digital integrated circuits implies obtaining sufficient timing performance from the desing. This may mean, for example, being able to operate the clock fast enough to obtain the required performance while guaranteeing functional correctness. Achieving timing closure is an important, iterative and time-consuming step in the design of any digital integrated circuit. Particularly in microprocessor designs, timing requirements, logic requirements and technology parameters are often changed late in the design cycle, making automated design closure techniques extremely valuable.
- Prior-art methods are illutstrated in FIG. 1 (flow 100 ). Because the overall design is too large and complex to optimize at once, prior-art methods typically divide the design into partitions called macros, and assign to each individual macro a timing and area budget by aprocess of apportionment (box 110 ). Then each macro is designed or the design is refined with the goal of meeting its budget, either by a process of automated synthesis or by means of custom design techniques (box 120 ). The optimization at this stage takes many forms such as logic re-structuring, buffer insertion, transistor sizing and use of low threshold voltage devices. The resulting design is timed, typically by means of static timing analysis (box 130 ).
- timing closure is achieved and the design is complete (box 150 ). More typically, the apportionment process is imperfect and involves some conjecture and guesswork. Hence, several macros will not meet their budgets, and overall timing closure is not achieved, as detected by box 140 . In this case, the apportionment process is repeated (box 110 ), individual macros are then redesigned and/or re-optimized (box 120 ), and the resulting overall design is timed (box 130 ), and this process iteratively repeated until timing closure is obtained (box 150 ), as depicted in FIG. 1 .
- FIG. 2 The main difficulty in prior-art techniques is that the application of automatic optimization techniques on individual macros interferes with the achievement of overall timing closure. This problem is illustrated in FIG. 2 .
- a short path of delay 200 time units of macro A feeds a long path of delay 600 time units of macro B.
- a different long path of delay 600 time units of macro A feeds a different short path of delay 200 time units of macro B, as shown in FIG. 2 .
- all output signals are required to be available by time 700.
- the initial design is missing timing closure by 100 time units, or, in other words, the initial design has a slack of ⁇ 100 time units.
- Slack is defined as the algebraic difference between required arrival time (RAT) and actual arrival time (AT).
- RAT required arrival time
- AT actual arrival time
- One particular prior-art apportionment technique will assign this negative slack of 100 time units to each of the two macros, giving the optimization procedures applied to each macro the opportunity to see and correct the entire negative slack of the global path.
- the required arrival times will be 100 and 500 at the upper and lower outputs of macro A, respectively, and 700 at both the upper and lower outputs of macro B, and the arrival times will be zero at both the upper and lower inputs of macro A, and 200 and 600 at the upper and lower inputs of macro B, respectively, as shown in the Figure.
- FIG. 3 Another prior-art apportionment method, one iteration of which is illustrated in FIG. 3 , would divide the negative slack according to the fraction of the global path delay suffered in each macro, and in the example of FIG. 2 , would assign ⁇ 25 time units of the upper path slack to macro A, ⁇ 75 of the upper path slack to macro B, ⁇ 75 of the lower path slack to macro A, and ⁇ 25 of the lower path slack to macro B.
- the required arrival times will be 175 and 525 at the upper and lower outputs of macro A, respectively, and 700 and 700 at both the upper and lower outputs of the second macro B, and the arrival times will be zero at both the upper and lower inputs of macro A, and 175 and 525 at the upper and lower inputs of macro B, respectively.
- the situation after one iteration is depicted in FIG. 3 .
- each of the delays through each of the macros can be decreased by 50 units by optimization.
- prior-art methods will never achieve timing closure under this apportionment scheme, since the redesign and re-optimization of individual macros typically target the worst slack, and because the long paths cannot be improved beyond 550, the redesign and re-optimization techniques have no incentive to improve the delay of the short paths, and upon successive iterations through loop of FIG. 1 , the delays and targets will be adjusted by decreasing amounts, and will asmyptotically approach but not reach timing closure.
- the crux of the problem is that prior-art optimization techniques target only paths with the worst slack and therefore do not improve sub-critical slacks even though such actions would help achieve timing closure from a global vantage point. Improviding sub-critical paths also makes it easier downstream in the methodology to focus design efforts in limited areas of the circuit to obtain timing convergence. Thus the formulation of the objective function during individual macro optimization has the unwanted consequence of preventing or impeding overall timing convergence.
- This invention relates to an improved method for achieving timing closure.
- this invention reformulates the objective of the redesign and re-optimization phase so that there is an incentive during automatic optimization to improve not only the arrival time of the most critical signals, but other sub-critical signals as well.
- the inventive method creates an incentive to optimize the arrival time of every output signal, the incentive being proportional to the criticality of the signal.
- FIG. 1 illustrates a typical prior-art iterative procedure for achieving timing closure of a high-performace digital integrated circuit or functional unit of a high-performance idigital integrated circuit.
- FIG. 2 illustrates an example situation in which prior-art optimization techniques will lead to inefficient achievement of timing closure, or not achieve timing closure at all.
- FIG. 3 illustrated the same example situation as FIG. 2 , but with application of one iteration of a second prior-art apportionment scheme; again, prior-art optimization techniques will not achieve timing closure.
- FIG. 4 illustrates the dependence of the penalty contributed by a signal on its criticality, using the preferred objective function reformulation.
- FIG. 5 illustrate the dependence of the logarithm of the penalty contributed by a signal on its criticality, using the preferred objective function reformulation.
- FIG. 6 illustrates an example slack histogram showing the slack histogram before optimization, after optimization with prior-art formulation of the objective function and after optimization with the inventive objective function reformulation.
- the heart of the invetnion is the reformulation of the objective function of any optimization that is performed by automatic means to improve individual macros.
- the reformulation makes the overall timing closure loop more effective and efficient.
- Traditional circuit optimization tools such as EinsTuner formulate the optimization problem in one of two ways, as described below.
- a description of EinsTuner is available in A. R. Conn, I. N. Elfadel, W. W. Molzen, Jr., P. R. O'Brien, P. N. Strenski, C. Visweswariah and C. B. Whan, “Gradient-based optimization of custom circuits using a static-timing formulation,” Proceedings of the 1999 Design Automation Conference, June 1999, pages 452-459.
- the description below assumes a simple combinational circuit to illustate the detailed description of the invention. It is to be understood that the formulation can easily be extended to situations containing sequential elements including all types of latches and possibly multiple clocks by one of skill in the art.
- the reformulation of the objective function is being demonstrated by means of the example of transistor sizing by a formal, mathematical optimizer.
- inventive method is applicable to any type of circuit change such as logic restructuring, buffering or use of low threshold voltage devices, and to any optimization method such as heuristic optimization, linear programming, nonlinear programming, branch-and-bound, dynamic programming or simulated annealing, provided the method of optimization makes use of an objective (or “cost” or “merit”) function that is to be minimized.
- the inventive method simple reformulates that objective function.
- z is larger than the negative of the worst slack among all the primary otuputs, and has the smallest possible value, hence the circuit has the smallest possible negative slack, or equivalently, the largest possible positive slack. It is clear that such a formulation will lead to a large number of equally critical paths, as explained in the above-mentioned Design Automation reference. Further, it is clear that if there is a limiting signal whose slack cannot be further improved, optimization based on this prior-art formulation has no incentive to improve any signal with a slack worse than the limiting signal's slack.
- desired_slack merely provides a notational convenience, since the required arrival times could be modified to reflect the additional desired slack. Note that a positive required_slack value makes the problem more difficult to solve.
- f is a penalty function and slack i is the effective slack of the i th primary output, taking desired_slack itno account.
- f is a penalty function
- slack i is the effective slack of the i th primary output, taking desired_slack itno account.
- the negative slack of each and every primary output is represented in the objective function.
- the key decision to be made is the choice of the function f, since it is desired that the signals that are most critical contribute the msot to the objective function, thus giving the optimizer the most to the objective function, thus giving the optimizer the most incentive to improve the timing of such signals. At the same time, if those signals cannot be improved any more, it is desired that sub-critical signals also have substantial cnotribution to the objective function, thus incenting the optimizer to improve their timing properties as well.
- f should be a decreasing function of its argument. If applied in a formal mathematical continuous optimizer, f should
- the limiting primary output will have a slack equal to the worst_starting_slack (usually a negative number), and hence the contribution of this signal to the objective function is exp(8).
- the effective slack is 0, hence the contribution to the objective function is exp(3). If the timing of this signal further improves, the contribution to the objective function gets smaller, and the rate of decrease in the contribution to the objective function per unit of timing improvement also decreases. In the meanwhile, even if a signal does not achieve its timing requirements, there is sufficient incentive on sub-critical signals to improve their timing characteristics, since every signal contributes a term to the objective function. That term gets smaller as timing requirements are closer to being met.
- FIG. 5 shows the variation of f with its argument (negative of the effective slack)
- FIG. 6 shows the same data on a logarithmic scale.
- Typical results obtained by using this nivention on a sample individual macro are shown in the slack histogram of FIG. 6 .
- the horizontal axis represents slack
- the vertical axis value of a point on the curve represents the number of paths with that slack or better. It is clear that traditional optimization has no incentive to improve sub-critical paths.
- the inventive method not only improves the limiting slack of the macro, it also improves the timing characteristics of each and every signal with an incentive proportional to its respective criticality.
- the objective function in this invention is re forumlated to obtain separation while simultaneously incenting the optimizer to work hardest on the most critical path(s) in order to enhance timing convergence at the functional-unit or chip-level.
- the uncertainty-awareness and all the benefits thereof that are obtained by increased separation are preserved by the present invention.
- the commercial application of our invetnion can be applied to any type of formal or heuristic optimization which requires the formulation of an objective function; it can be applied to any type of circuitry that is amenable to static timing analysis; it can be extended to application having master-slave latches, transparent latches, multi-cycle clocks, multi-frequency clocks and dynamic circuits.
Abstract
Description
- This is a divisional of U.S. Ser. No. 10/435,824, filed May 12, 2003, by Jun Zhou et al.
- This invention relates to the design, and automation thereof, of high-performance digital integrated circuits. The invention is particularly directed to the problem of obtaining timing closure of entire integrated circuits or functional units of an integrated circuit by optimizing or tuning individual macros that contitute the functional unit or integrated circuit.
- D. J. Hathaway, L. K. Lange, C. Visweswariah and P. M. Williams, “Method of Optimizing and Analyzing Selected Portions of a Digital Integrated Circuit,” filed May 12, 2003 under U.S. Ser. No. 10/426,213.
- E. K. Cho, D. J. Hathaway, M. Hsu, L. K. Lange, G. a. Northrop, C. Visweswariah, C. Washburn, P. J. Williams, J. Zhou, “A Method for Tuning a Digital Design for Synthesized Random Logic Circuit Macros in a Continuous Design Space with Optimal Insertion of Multiple Threshold Voltage Devices” filed May 12, 2003 under U.S. Ser. No. 10/42,589.
- These co-pending applications and the present application are owned by one and the same assignee, International Business Machines Corporation of Armonk, N.Y.
- The descriptions set forth in these co-pending applications are hereby incorporated into the present application by this reference.
- Trademarks: IBM® is a registered trademark of International Business Machines Corporation, Armonk, N.Y., U.S.A. Other names may be registered trademarks or product names of International Business Machines Corporation or other companies.
- Achieving timing closure of high-performace digital integrated circuits (or functional units of a high-performance digital integrated circuit) implies obtaining sufficient timing performance from the desing. This may mean, for example, being able to operate the clock fast enough to obtain the required performance while guaranteeing functional correctness. Achieving timing closure is an important, iterative and time-consuming step in the design of any digital integrated circuit. Particularly in microprocessor designs, timing requirements, logic requirements and technology parameters are often changed late in the design cycle, making automated design closure techniques extremely valuable.
- Prior-art methods are illutstrated in
FIG. 1 (flow 100). Because the overall design is too large and complex to optimize at once, prior-art methods typically divide the design into partitions called macros, and assign to each individual macro a timing and area budget by aprocess of apportionment (box 110). Then each macro is designed or the design is refined with the goal of meeting its budget, either by a process of automated synthesis or by means of custom design techniques (box 120). The optimization at this stage takes many forms such as logic re-structuring, buffer insertion, transistor sizing and use of low threshold voltage devices. The resulting design is timed, typically by means of static timing analysis (box 130). If every macro meets its budget, it is obvious that timing closure is achieved and the design is complete (box 150). More typically, the apportionment process is imperfect and involves some conjecture and guesswork. Hence, several macros will not meet their budgets, and overall timing closure is not achieved, as detected bybox 140. In this case, the apportionment process is repeated (box 110), individual macros are then redesigned and/or re-optimized (box 120), and the resulting overall design is timed (box 130), and this process iteratively repeated until timing closure is obtained (box 150), as depicted inFIG. 1 . - The main difficulty in prior-art techniques is that the application of automatic optimization techniques on individual macros interferes with the achievement of overall timing closure. This problem is illustrated in
FIG. 2 . Consider the simple case of macro A (box 200) feeding macro B (box 210). A short path ofdelay 200 time units of macro A feeds a long path ofdelay 600 time units of macro B. A different long path ofdelay 600 time units of macro A feeds a different short path ofdelay 200 time units of macro B, as shown inFIG. 2 . Assume that all output signals are required to be available bytime 700. In this case, the initial design is missing timing closure by 100 time units, or, in other words, the initial design has a slack of −100 time units. Slack is defined as the algebraic difference between required arrival time (RAT) and actual arrival time (AT). One particular prior-art apportionment technique will assign this negative slack of 100 time units to each of the two macros, giving the optimization procedures applied to each macro the opportunity to see and correct the entire negative slack of the global path. Using this apportionment method, the required arrival times will be 100 and 500 at the upper and lower outputs of macro A, respectively, and 700 at both the upper and lower outputs of macro B, and the arrival times will be zero at both the upper and lower inputs of macro A, and 200 and 600 at the upper and lower inputs of macro B, respectively, as shown in the Figure. - Suppose the short paths cannot be improved, but there is room for improvement int he long paths. It is clear from this example that improving the two long paths from 600 to 500 units will achieve overall timing closure. Unfortunately, prior-art methods will never achieve timing closure in this case, since the redesign and re-optimization of individual macros typically target the worst clask, and because the short paths cannot be improved, the redesign and re-optimization techniques have no incentive to improve the delay of the long paths.
- Another prior-art apportionment method, one iteration of which is illustrated in
FIG. 3 , would divide the negative slack according to the fraction of the global path delay suffered in each macro, and in the example ofFIG. 2 , would assign −25 time units of the upper path slack to macro A, −75 of the upper path slack to macro B, −75 of the lower path slack to macro A, and −25 of the lower path slack to macro B. Using this approtionment method, the required arrival times will be 175 and 525 at the upper and lower outputs of macro A, respectively, and 700 and 700 at both the upper and lower outputs of the second macro B, and the arrival times will be zero at both the upper and lower inputs of macro A, and 175 and 525 at the upper and lower inputs of macro B, respectively. The situation after one iteration is depicted inFIG. 3 . - Suppose now that each of the delays through each of the macros can be decreased by 50 units by optimization. Again, prior-art methods will never achieve timing closure under this apportionment scheme, since the redesign and re-optimization of individual macros typically target the worst slack, and because the long paths cannot be improved beyond 550, the redesign and re-optimization techniques have no incentive to improve the delay of the short paths, and upon successive iterations through loop of
FIG. 1 , the delays and targets will be adjusted by decreasing amounts, and will asmyptotically approach but not reach timing closure. - With this second prior-art apportionment method, if the long paths in each macro can be improved by 100 units each, and the short paths cannot be improved at all, it is clear that although an easy solution exists for global timing closure, the iteration of
FIG. 1 will not converge to the solution in reasonable time. The reason is that the short path's stubborn negative slack at each iteration ofFIG. 1 will limit the improvement that is targeted for the long path of each macro. - Irrespective of the apportionment method employed, the crux of the problem is that prior-art optimization techniques target only paths with the worst slack and therefore do not improve sub-critical slacks even though such actions would help achieve timing closure from a global vantage point. Improviding sub-critical paths also makes it easier downstream in the methodology to focus design efforts in limited areas of the circuit to obtain timing convergence. Thus the formulation of the objective function during individual macro optimization has the unwanted consequence of preventing or impeding overall timing convergence.
- It is to be appreciated that this simple example merely illustrates the problem. With a large number of macros and a large number of interconnections between them, the problem is exacerbated and achievement of timing closure becomes an extremely hard proboem, leading to costly redesign efforts and increased time-to-market of the product.
- This invention relates to an improved method for achieving timing closure. During the design iterations, focusing solely on the most critical (or limiting) slack inhibits overall timing closure. Instead, this invention reformulates the objective of the redesign and re-optimization phase so that there is an incentive during automatic optimization to improve not only the arrival time of the most critical signals, but other sub-critical signals as well. Instead of the prior-art focus on the most critical signal or signals, the inventive method creates an incentive to optimize the arrival time of every output signal, the incentive being proportional to the criticality of the signal. Thus once the most critical signals cannot be further improved, sub-critical signals are improved, leading to more efficient and effective overal timing closure.
- These and other improvements are set forth in the following detailed description. For a better understanding of the invention with advantages and features, please refer to the detailed description and to the drawings.
-
FIG. 1 illustrates a typical prior-art iterative procedure for achieving timing closure of a high-performace digital integrated circuit or functional unit of a high-performance idigital integrated circuit. -
FIG. 2 illustrates an example situation in which prior-art optimization techniques will lead to inefficient achievement of timing closure, or not achieve timing closure at all. -
FIG. 3 illustrated the same example situation asFIG. 2 , but with application of one iteration of a second prior-art apportionment scheme; again, prior-art optimization techniques will not achieve timing closure. -
FIG. 4 illustrates the dependence of the penalty contributed by a signal on its criticality, using the preferred objective function reformulation. -
FIG. 5 illustrate the dependence of the logarithm of the penalty contributed by a signal on its criticality, using the preferred objective function reformulation. -
FIG. 6 illustrates an example slack histogram showing the slack histogram before optimization, after optimization with prior-art formulation of the objective function and after optimization with the inventive objective function reformulation. - Our detailed description below explains the preferred embodiments of our invention, together with advantages nd features, by way of example with reference to the drawings.
- The heart of the invetnion is the reformulation of the objective function of any optimization that is performed by automatic means to improve individual macros. The reformulation makes the overall timing closure loop more effective and efficient. Traditional circuit optimization tools such as EinsTuner formulate the optimization problem in one of two ways, as described below. A description of EinsTuner is available in A. R. Conn, I. N. Elfadel, W. W. Molzen, Jr., P. R. O'Brien, P. N. Strenski, C. Visweswariah and C. B. Whan, “Gradient-based optimization of custom circuits using a static-timing formulation,” Proceedings of the 1999 Design Automation Conference, June 1999, pages 452-459. The description below assumes a simple combinational circuit to illustate the detailed description of the invention. It is to be understood that the formulation can easily be extended to situations containing sequential elements including all types of latches and possibly multiple clocks by one of skill in the art.
- It is to be noted that the reformulation of the objective function is being demonstrated by means of the example of transistor sizing by a formal, mathematical optimizer. However, the inventive method is applicable to any type of circuit change such as logic restructuring, buffering or use of low threshold voltage devices, and to any optimization method such as heuristic optimization, linear programming, nonlinear programming, branch-and-bound, dynamic programming or simulated annealing, provided the method of optimization makes use of an objective (or “cost” or “merit”) function that is to be minimized. The inventive method simple reformulates that objective function.
- The first traditional formulation is delay minimization in which the problem is formulated as follows:
min z
s.t.≧AT i −RAT i ,i=1,2, . . . ,n
where z is an auxiliary optimization variable representing the negative slack of the circuit, n is the number of primary output signals of the combinational circuit, Ati and RATi are the arrival time and required arrival time of the ith primary otuput signal. It is to be understood that many other constraints like area and slew constraints are required to render the results of the optimization practical, but the simplistic formulation above serves to illustrate a point. At optimality, z is larger than the negative of the worst slack among all the primary otuputs, and has the smallest possible value, hence the circuit has the smallest possible negative slack, or equivalently, the largest possible positive slack. It is clear that such a formulation will lead to a large number of equally critical paths, as explained in the above-mentioned Design Automation reference. Further, it is clear that if there is a limiting signal whose slack cannot be further improved, optimization based on this prior-art formulation has no incentive to improve any signal with a slack worse than the limiting signal's slack. - The second traditional formulation is area minimization, in which the problem is formulated as follows:
min area
s.t. At i ≦RAT i−desired_slack,i=1,2, . . . ,n
where the area of the circuit is minimized subject to tiing constraints, and desired_slack represents an (algebraic) additional slack required by the user. Using desired_slack merely provides a notational convenience, since the required arrival times could be modified to reflect the additional desired slack. Note that a positive required_slack value makes the problem more difficult to solve. It is clear that even in this second formulation, a large number of equally critical paths will result, especially since area is “stolen” from sub-critical paths to speed up critical paths. Further, once a primary output signal achieves its timing requirement, there is no further incentive to improve its timing. - Thus, both traditional optimization formulations described above do not solve the problem of encouraging the optimizer to pay attention to sub-critical paths. Instead, this invention proposes a new formulation of the objective function as follows:
- where f is a penalty function and slacki is the effective slack of the ith primary output, taking desired_slack itno account. Thus the negative slack of each and every primary output is represented in the objective function. The key decision to be made is the choice of the function f, since it is desired that the signals that are most critical contribute the msot to the objective function, thus giving the optimizer the most to the objective function, thus giving the optimizer the most incentive to improve the timing of such signals. At the same time, if those signals cannot be improved any more, it is desired that sub-critical signals also have substantial cnotribution to the objective function, thus incenting the optimizer to improve their timing properties as well. Clearly, f should be a decreasing function of its argument. If applied in a formal mathematical continuous optimizer, f should preferably be a smooth, continuous and continuously differentiable function.
- In a preferred version of this ivnention, the choice of f is as follows:
- At the start of the optimization, the limiting primary output will have a slack equal to the worst_starting_slack (usually a negative number), and hence the contribution of this signal to the objective function is exp(8). As the optimization progresses, if a primary output signal achieves its timing requirement, the effective slack is 0, hence the contribution to the objective functio is exp(3). If the timing of this signal further improves, the contribution to the objective function gets smaller, and the rate of decrease in the contribution to the objective function per unit of timing improvement also decreases. In the meanwhile, even if a signal does not achieve its timing requirements, there is sufficient incentive on sub-critical signals to improve their timing characteristics, since every signal contributes a term to the objective function. That term gets smaller as timing requirements are closer to being met.
FIG. 5 shows the variation of f with its argument (negative of the effective slack), andFIG. 6 shows the same data on a logarithmic scale. - Applying this method to the simple example of
FIG. 2 , we see that even thoughh the “short path” cannot be improved in the two individual macros being tuned, there is sufficient incentive to improve the long paths even though they are non-limiting paths, and the overall lop moves towards timing closure. - Typical results obtained by using this nivention on a sample individual macro are shown in the slack histogram of
FIG. 6 . In a slack histogram, the horizontal axis represents slack, and the vertical axis value of a point on the curve represents the number of paths with that slack or better. It is clear that traditional optimization has no incentive to improve sub-critical paths. The inventive method not only improves the limiting slack of the macro, it also improves the timing characteristics of each and every signal with an incentive proportional to its respective criticality. - One reason to reformulate the objective function is to obtain more “separation,” where separation is the difference between the slack of a sub-critical path and the overall slack of the macro. This separation has beneficial properties in obtaining global timing convergence, as taught by this invention. It is to be noted that another reason to obtain separation is to be relatively immune to downstream changes in the design, modeling or manufacturing of the circuit, as disclosed in X. Bai, D. J. Hathaway, P. N. Strenki, and C. Visweswariah, “Parameter-Variation Tlerant Method for Circuit Design Optimization,” Docket FIS920020034US1, filed May 30, 2002. In that invention, penalty terms were added to the traditional objective function to obtain separation in order to be tolerant to downstream changes or modeling uncertainties. In contrast, the objective function in this invention is reforumlated to obtain separation while simultaneously incenting the optimizer to work hardest on the most critical path(s) in order to enhance timing convergence at the functional-unit or chip-level. The uncertainty-awareness and all the benefits thereof that are obtained by increased separation are preserved by the present invention.
- The commercial application of our invetnion can be applied to any type of formal or heuristic optimization which requires the formulation of an objective function; it can be applied to any type of circuitry that is amenable to static timing analysis; it can be extended to application having master-slave latches, transparent latches, multi-cycle clocks, multi-frequency clocks and dynamic circuits.
- While the preferred embodiment ot the invention has been described, it will be understood that those skilled in the art, both now and in the future, may make various improvements and enhancements which fall within the scope of the claims which follow. These claims should be construed to maintain the proper protection for the invention first described.
Claims (13)
Priority Applications (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US11/296,780 US20060150127A1 (en) | 2003-05-12 | 2005-12-07 | Method of achieving timing closure in digital integrated circuits by optimizing individual macros |
US11/942,034 US7743355B2 (en) | 2003-05-12 | 2007-11-19 | Method of achieving timing closure in digital integrated circuits by optimizing individual macros |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US10/435,824 US7003747B2 (en) | 2003-05-12 | 2003-05-12 | Method of achieving timing closure in digital integrated circuits by optimizing individual macros |
US11/296,780 US20060150127A1 (en) | 2003-05-12 | 2005-12-07 | Method of achieving timing closure in digital integrated circuits by optimizing individual macros |
Related Parent Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
US10/435,824 Division US7003747B2 (en) | 2003-05-12 | 2003-05-12 | Method of achieving timing closure in digital integrated circuits by optimizing individual macros |
Related Child Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
US11/942,034 Division US7743355B2 (en) | 2003-05-12 | 2007-11-19 | Method of achieving timing closure in digital integrated circuits by optimizing individual macros |
Publications (1)
Publication Number | Publication Date |
---|---|
US20060150127A1 true US20060150127A1 (en) | 2006-07-06 |
Family
ID=33417013
Family Applications (3)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
US10/435,824 Expired - Fee Related US7003747B2 (en) | 2003-05-12 | 2003-05-12 | Method of achieving timing closure in digital integrated circuits by optimizing individual macros |
US11/296,780 Abandoned US20060150127A1 (en) | 2003-05-12 | 2005-12-07 | Method of achieving timing closure in digital integrated circuits by optimizing individual macros |
US11/942,034 Expired - Fee Related US7743355B2 (en) | 2003-05-12 | 2007-11-19 | Method of achieving timing closure in digital integrated circuits by optimizing individual macros |
Family Applications Before (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
US10/435,824 Expired - Fee Related US7003747B2 (en) | 2003-05-12 | 2003-05-12 | Method of achieving timing closure in digital integrated circuits by optimizing individual macros |
Family Applications After (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
US11/942,034 Expired - Fee Related US7743355B2 (en) | 2003-05-12 | 2007-11-19 | Method of achieving timing closure in digital integrated circuits by optimizing individual macros |
Country Status (1)
Country | Link |
---|---|
US (3) | US7003747B2 (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7207020B1 (en) * | 2004-02-09 | 2007-04-17 | Altera Corporation | Method and apparatus for utilizing long-path and short-path timing constraints in an electronic-design-automation tool |
US8156463B1 (en) | 2004-02-09 | 2012-04-10 | Altera Corporation | Method and apparatus for utilizing long-path and short-path timing constraints in an electronic-design-automation tool for routing |
Families Citing this family (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7003747B2 (en) * | 2003-05-12 | 2006-02-21 | International Business Machines Corporation | Method of achieving timing closure in digital integrated circuits by optimizing individual macros |
US7093208B2 (en) * | 2003-05-12 | 2006-08-15 | International Business Machines Corporation | Method for tuning a digital design for synthesized random logic circuit macros in a continuous design space with optional insertion of multiple threshold voltage devices |
KR20040103673A (en) * | 2003-06-02 | 2004-12-09 | 삼성전자주식회사 | Flat panel display device |
US7117466B2 (en) * | 2003-09-18 | 2006-10-03 | International Business Machines Corporation | System and method for correlated process pessimism removal for static timing analysis |
US7155692B2 (en) * | 2004-09-01 | 2006-12-26 | Hewlett-Packard Development Company, L.P. | Method and system for performing timing analysis on a circuit |
US20080307374A1 (en) * | 2007-06-05 | 2008-12-11 | International Business Machines Corporation | Method, system, and computer program product for mapping a logical design onto an integrated circuit with slack apportionment |
US8595674B2 (en) * | 2007-07-23 | 2013-11-26 | Synopsys, Inc. | Architectural physical synthesis |
US8819608B2 (en) * | 2007-07-23 | 2014-08-26 | Synopsys, Inc. | Architectural physical synthesis |
US7853912B2 (en) * | 2007-11-05 | 2010-12-14 | International Business Machines Corporation | Arrangements for developing integrated circuit designs |
US7836418B2 (en) * | 2008-03-24 | 2010-11-16 | International Business Machines Corporation | Method and system for achieving power optimization in a hierarchical netlist |
US8307315B2 (en) * | 2009-01-30 | 2012-11-06 | Synopsys, Inc. | Methods and apparatuses for circuit design and optimization |
US8516417B2 (en) | 2009-08-07 | 2013-08-20 | International Business Machines Corporation | Method and system for repartitioning a hierarchical circuit design |
US9298872B2 (en) | 2014-02-20 | 2016-03-29 | International Business Machines Corporation | Apportioning synthesis effort for better timing closure |
US9177096B2 (en) | 2014-03-26 | 2015-11-03 | Freescale Semiconductor, Inc. | Timing closure using transistor sizing in standard cells |
US10568203B2 (en) * | 2017-06-07 | 2020-02-18 | International Business Machines Corporation | Modifying a circuit design |
Citations (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US4495559A (en) * | 1981-11-02 | 1985-01-22 | International Business Machines Corporation | Optimization of an organization of many discrete elements |
US4698760A (en) * | 1985-06-06 | 1987-10-06 | International Business Machines | Method of optimizing signal timing delays and power consumption in LSI circuits |
US5237514A (en) * | 1990-12-21 | 1993-08-17 | International Business Machines Corporation | Minimizing path delay in a machine by compensation of timing through selective placement and partitioning |
US20010010090A1 (en) * | 1998-02-11 | 2001-07-26 | Boyle Douglas B. | Method for design optimization using logical and physical information |
US6327552B2 (en) * | 1999-12-28 | 2001-12-04 | Intel Corporation | Method and system for determining optimal delay allocation to datapath blocks based on area-delay and power-delay curves |
US6408428B1 (en) * | 1999-08-20 | 2002-06-18 | Hewlett-Packard Company | Automated design of processor systems using feedback from internal measurements of candidate systems |
US6480991B1 (en) * | 2001-04-11 | 2002-11-12 | International Business Machines Corporation | Timing-driven global placement based on geometry-aware timing budgets |
US6618849B2 (en) * | 2001-08-23 | 2003-09-09 | Cadence Design Systems, Inc. | Method and apparatus for identifying routes for nets |
US6766504B1 (en) * | 2002-08-06 | 2004-07-20 | Xilinx, Inc. | Interconnect routing using logic levels |
US6826733B2 (en) * | 2002-05-30 | 2004-11-30 | International Business Machines Corporation | Parameter variation tolerant method for circuit design optimization |
US7003747B2 (en) * | 2003-05-12 | 2006-02-21 | International Business Machines Corporation | Method of achieving timing closure in digital integrated circuits by optimizing individual macros |
Family Cites Families (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US449559A (en) * | 1891-03-31 | Gar-coupling | ||
US5218551A (en) * | 1990-04-30 | 1993-06-08 | International Business Machines Corporation | Timing driven placement |
US6591407B1 (en) * | 2000-03-01 | 2003-07-08 | Sequence Design, Inc. | Method and apparatus for interconnect-driven optimization of integrated circuit design |
US6717921B1 (en) * | 2000-05-17 | 2004-04-06 | Lucent Technologies Inc. | Method for configuring a shared tree for routing traffic in a multicast conference |
US6721924B2 (en) * | 2001-09-28 | 2004-04-13 | Intel Corporation | Noise and power optimization in high performance circuits |
US6765981B2 (en) * | 2002-07-31 | 2004-07-20 | Agilent Technologies, Inc. | Computed tomography |
US7111268B1 (en) * | 2002-08-26 | 2006-09-19 | Xilinx, Inc. | Post-layout optimization in integrated circuit design |
US6954909B2 (en) * | 2003-02-12 | 2005-10-11 | Stmicroelectronics, Inc. | Method for synthesizing domino logic circuits |
US7251800B2 (en) * | 2003-05-30 | 2007-07-31 | Synplicity, Inc. | Method and apparatus for automated circuit design |
-
2003
- 2003-05-12 US US10/435,824 patent/US7003747B2/en not_active Expired - Fee Related
-
2005
- 2005-12-07 US US11/296,780 patent/US20060150127A1/en not_active Abandoned
-
2007
- 2007-11-19 US US11/942,034 patent/US7743355B2/en not_active Expired - Fee Related
Patent Citations (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US4495559A (en) * | 1981-11-02 | 1985-01-22 | International Business Machines Corporation | Optimization of an organization of many discrete elements |
US4698760A (en) * | 1985-06-06 | 1987-10-06 | International Business Machines | Method of optimizing signal timing delays and power consumption in LSI circuits |
US5237514A (en) * | 1990-12-21 | 1993-08-17 | International Business Machines Corporation | Minimizing path delay in a machine by compensation of timing through selective placement and partitioning |
US20010010090A1 (en) * | 1998-02-11 | 2001-07-26 | Boyle Douglas B. | Method for design optimization using logical and physical information |
US6408428B1 (en) * | 1999-08-20 | 2002-06-18 | Hewlett-Packard Company | Automated design of processor systems using feedback from internal measurements of candidate systems |
US6327552B2 (en) * | 1999-12-28 | 2001-12-04 | Intel Corporation | Method and system for determining optimal delay allocation to datapath blocks based on area-delay and power-delay curves |
US6480991B1 (en) * | 2001-04-11 | 2002-11-12 | International Business Machines Corporation | Timing-driven global placement based on geometry-aware timing budgets |
US6618849B2 (en) * | 2001-08-23 | 2003-09-09 | Cadence Design Systems, Inc. | Method and apparatus for identifying routes for nets |
US6826733B2 (en) * | 2002-05-30 | 2004-11-30 | International Business Machines Corporation | Parameter variation tolerant method for circuit design optimization |
US6766504B1 (en) * | 2002-08-06 | 2004-07-20 | Xilinx, Inc. | Interconnect routing using logic levels |
US7003747B2 (en) * | 2003-05-12 | 2006-02-21 | International Business Machines Corporation | Method of achieving timing closure in digital integrated circuits by optimizing individual macros |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US7207020B1 (en) * | 2004-02-09 | 2007-04-17 | Altera Corporation | Method and apparatus for utilizing long-path and short-path timing constraints in an electronic-design-automation tool |
US8156463B1 (en) | 2004-02-09 | 2012-04-10 | Altera Corporation | Method and apparatus for utilizing long-path and short-path timing constraints in an electronic-design-automation tool for routing |
Also Published As
Publication number | Publication date |
---|---|
US7003747B2 (en) | 2006-02-21 |
US20080072184A1 (en) | 2008-03-20 |
US7743355B2 (en) | 2010-06-22 |
US20040230929A1 (en) | 2004-11-18 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US7743355B2 (en) | Method of achieving timing closure in digital integrated circuits by optimizing individual macros | |
Choi et al. | Novel sizing algorithm for yield improvement under process variation in nanometer technology | |
US7093208B2 (en) | Method for tuning a digital design for synthesized random logic circuit macros in a continuous design space with optional insertion of multiple threshold voltage devices | |
Bai et al. | Uncertainty-aware circuit optimization | |
Davare et al. | The best of both worlds: The efficient asynchronous implementation of synchronous specifications | |
US7360198B2 (en) | Technology dependent transformations for CMOS in digital design synthesis | |
US7761826B1 (en) | Method and system for crosstalk analysis | |
US5956257A (en) | Automated optimization of hierarchical netlists | |
CN100350414C (en) | System and method for statistical timing analysis of digital circuits | |
US7010763B2 (en) | Method of optimizing and analyzing selected portions of a digital integrated circuit | |
US5553000A (en) | Eliminating retiming bottlenecks to improve performance of synchronous sequential VLSI circuits | |
US20120284680A1 (en) | Method and apparatus for designing an integrated circuit | |
Ababei et al. | Multi-objective circuit partitioning for cutsize and path-based delay minimization | |
US20100070941A1 (en) | Achieving Clock Timing Closure in Designing an Integrated Circuit | |
Zhu et al. | Optimal sizing of high-speed clock networks based on distributed RC and lossy transmission line models | |
US5963730A (en) | Method for automating top-down design processing for the design of LSI functions and LSI mask layouts | |
US6086621A (en) | Logic synthesis constraints allocation automating the concurrent engineering flows | |
Tennakoon et al. | Efficient and accurate gate sizing with piecewise convex delay models | |
US6526541B2 (en) | Library for use in designing a semiconductor device | |
Kahng et al. | Placement feedback: A concept and method for better min-cut placements | |
Wang et al. | A variation aware high level synthesis framework | |
US6321362B1 (en) | Method of reformulating static circuit optimization problems for reduced size, degeneracy and redundancy | |
Lucas et al. | FastYield: Variation-aware, layout-driven simultaneous binding and module selection for performance yield optimization | |
Yang et al. | A standard-cell placement tool for designs with high row utilization | |
Kang et al. | Statistical timing analysis using levelized covariance propagation considering systematic and random variations of process parameters |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
STCB | Information on status: application discontinuation |
Free format text: ABANDONED -- FAILURE TO RESPOND TO AN OFFICE ACTION |
|
AS | Assignment |
Owner name: GLOBALFOUNDRIES U.S. 2 LLC, NEW YORK Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNOR:INTERNATIONAL BUSINESS MACHINES CORPORATION;REEL/FRAME:036550/0001 Effective date: 20150629 |
|
AS | Assignment |
Owner name: GLOBALFOUNDRIES INC., CAYMAN ISLANDS Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNORS:GLOBALFOUNDRIES U.S. 2 LLC;GLOBALFOUNDRIES U.S. INC.;REEL/FRAME:036779/0001 Effective date: 20150910 |